T1-DWI Co-Registration Query

[LaraFernandez44]

Hi, My research group has recently used QSIprep, and would like some information on the DWI-to-T1 registration stage. While the majority of registrations were good, we noticed in a subset of images, a less than ideal alignment between the DWI and T1 volumes. To correct this, we re-did the co-registration in SPM using the default optimisation metric: normalised mutual information. Our understanding is that this is a histogram-based method, while QSIPrep uses ANT's SyN (mutual information). Is it possible to use a histogram-matching based approach in QSIPrep, similar to the metric used in SPM? For example, via: antsRegistrationSyN --use histogram matching 1?

Many thanks, Lara.

[mattcieslak]

Hi @LaraFernandez44, for T1w to DWI registration we use ANTs, but a rigid registration that actually does use mutual information (called "Mattes" in ANTs and is also histogram-based). The histogram matching option enables a signal preprocessing step that can be done outside of the registration. Although I would double-check this with @cookpa

[cookpa]

This is correct, the histogram-matching option for antsRegistration does a histogram match pre-processing, similar to what you could do with ImageMath.

The mutual information metric in antsRegistration is Mattes Mutual information. There are other MI implementations in ITK but they are not used in antsRegistration. The metric is only one parameter though, it's hard to say if the metric is what makes SPM work better in this case. It could be initialization, multi-resolution parameters, or something else.